Curated Cancer Cell Atlas (CCCA) Data Download Guide

Curated Cancer Cell Atlas (CCCA) Data Download Guide#

The CanSig-benchmark project provides functionality to download and process datasets from the Curated Cancer Cell Atlas (CCCA). This guide explains how to download datasets and generate processed samples for analysis.

Overview#

The CCCA fetcher (ccafetcher.py) handles:

  • Downloading raw data and metadata from Dropbox URLs

  • Processing ZIP files containing expression matrices

  • Converting data to AnnData format for analysis

  • Merging metadata with expression data

Dataset Metadata Structure#

Datasets are defined in CSV files (e.g., 3ca.csv) with the following columns:

  • Tissue: Tissue type (e.g., lung, breast, brain)

  • Title: Study author and year

  • Article: DOI/URL to the original publication

  • Data: Dropbox URL for the expression data (ZIP format)

  • Meta_data: Dropbox URL for the metadata (CSV format)

  • Disease: Disease type or condition

  • Technology: Sequencing technology (10x, SmartSeq2, etc.)

  • N_cells: Number of cells in the dataset

  • N_sampels: Number of samples in the dataset

Basic Usage#

1. Download Only (Fetch Mode)#

To download raw datasets without processing:

⚠️ Warning: This will download all datasets specified in the metadata file. Please subset them to the files needed!

python ccafetcher.py --metadata 3ca.csv --download-dir data/downloads --fetch

This will:

  • Create the download directory if it doesn’t exist

  • Download all ZIP files and metadata CSV files

  • Store files with hashed names based on their URLs

2. Process Single Dataset#

To download and convert a specific dataset (first row in the metadata) to AnnData format:

python ccafetcher.py --metadata 3ca.csv --download-dir data/downloads --dataset-dir data/raw --sample 0

Parameters:

  • --metadata: Path to the metadata CSV file

  • --download-dir: Directory to store raw downloaded files

  • --dataset-dir: Directory to store processed AnnData files

  • --sample: Index of the dataset to process (0-based)

  • --fetch: Download files if not already present

Output Format#

Processed datasets are saved as .h5ad files (AnnData format) in the specified dataset directory. The naming convention is:

  • Single dataset per ZIP: {Author}_{Tissue}.h5ad

  • Multiple datasets per ZIP: {Author}_{Tissue}_{Subdirectory}.h5ad

Data Structure#

Each processed AnnData object contains:

  • X: Expression matrix (sparse CSR format)

  • obs: Cell-level metadata (observations)

  • var: Gene-level metadata (variables)

  • uns: Unstructured metadata including count type

Changes made to the 3CA.#

We corrected a couple of errors in the 3CA.

  • Fix the metadata of Neftel et al. by changing the technology to 10x for some samples.

  • Converted the .rds in Couturier et al. to .mtx.

  • The dataset from Couturier et al. was subsetted to cells that where identified as IDH WT in genetic_hormonal_features and in histology as GBM.

  • The dataset from Yuan et al. was subsetted to GBM patients.

Contents